Qingfeng Hua

Accurate prediction of marine soil undrained shear strength is crucial for deep-sea engineering but is challenged by costly testing and the overfitting susceptibility of single machine learning models. This study proposes a Stacking ensemble framework to address these issues. First, a deep autoencoder cleaned outliers from Chinese sea data. Subsequently, hybrid data augmentation—incorporating a Gaussian Mixture Model, Correlation-Preserving Interpolation, and Boundary-Aware Noise Injection—expanded the dataset to 10,000 samples, preserving underlying physical distributions. A Stacking model was developed utilizing six heterogeneous base learners (including Random Forest, Gradient Boosting Machine, eXtreme Gradient Boosting, Light Gradient Boosting Machine, Support Vector Regression_linear and Support Vector Regression_rbf) and an L2-regularized ridge regression meta-learner. It achieved superior accuracy (R^2 = 0.961) and robustness. Shapley Additive Explanations resolved interpretability challenges, verifying these parameters as primary strength drivers with nonlinear responses consistent with classical soil mechanics. Streamlining inputs to five core indicators (shear wave velocity, moisture content, density, clay content, and void ratio) maintained high fidelity (R^2 = 0.920), another size parameters proving the application potential of acoustic detection and particle size data. This research provides a reliable, physically self-consistent method for the intelligent, non-destructive inversion of seabed strength in complex marine environments.

The acoustic properties of seafloor sediments on continental slopes play a crucial role in underwater acoustic propagation, communication, and detection. To investigate the acoustic characteristics and spatial distribution patterns of sediments on the continental slope, a geoacoustic experiment was conducted in the northwestern South China Sea. The experiment covered two sections: one crossing the shelf and slope in the downslope direction, and the other near the shelf break in the along-slope direction. In situ techniques, sediment sampling, and laboratory measurements were used to acquire data on sediment acoustic properties (such as sound speed and attenuation) and physical properties (including particle composition, density, porosity, and mean grain size). The experimental findings revealed several key points: (1) Acoustic properties of shallow water coarse-grained sediments and deep-sea sediments were higher when measured in the laboratory compared to in situ measurements. (2) Relationships between measured attenuation and physical properties, as well as between sound speed and mean grain size, showed deviations from previous empirical equations. (3) Sediment acoustic and physical properties exhibited significant variations in the downslope direction, while showing gradual variations in the along-slope direction. These variations can be attributed to sedimentary environmental factors such as material sources, hydrodynamic conditions, and water depth.

Based on the data of the acoustic and physical properties of seafloor sediments obtained on the continental shelf area of the East China Sea (ECS), prediction equations of the sediment sound speed based on single and dual parameters were established, and the correlation of the sound speed with physical parameters was discussed. The results show that the sediment sound speed (c) is strongly correlated with water content (w), density (ρ), void ratio (e), mean grain size (M_z) and median grain size (M_d), and the coefficient of determination R² of the empirical regression equation is generally greater than 0.80, while the empirical regression equation coefficient of determination R² of the compression coefficient (a) and compression modulus (E) are slightly lower, with 0.79 and 0.73, respectively. The coefficients of determination of the dual parameter regression equations between sediment sound speed with physical property parameters are generally higher than those of the single parameter equations, which are all higher than 0.90 indicating better prediction performance. The sensitivity of the physical parameters to the sound speed was analyzed, and the result shows that the sequence of sensitivity from high to low influence on sediment sound speed is void ratio, density, compressibility coefficient, median grain size and mean grain size. The prediction equations established in this study are a good extension and supplement to marine geoacoustic models and is of great significance for obtaining the acoustic properties of the seafloor sediments on the shelf of the East China Sea.

Based on multibeam bathymetric data and high-resolution shallow sub-bottom profiles acquired during China’s 10th Arctic Scientific Expedition Cruise in 2019, a sediment wave field was found on the western slope of the Chukchi Rise, in the Arctic Ocean. This sediment wave field developed on the lower slope with water depths of between 1200 m and 1800 m and stretched 15 km in the downslope direction. It comprised several parallel sediment waves, with wavelengths ranging from 700 m to 3400 m and wave heights from 12 m to 70 m. In the vertical direction, well-stratified deposits, tens of meters thick, were affected by sediment waves, which exhibit asymmetric waveforms and upslope migration trends. The morphological and seismostratigraphic characteristics of the sediment waves suggested their genesis as a result of the interaction between the bottom current and seafloor morphology, which was also supported by hydrographical data adjacent to the sediment wave field. It was infered that this bottom current was related to the Arctic Circumpolar Boundary Current, which many researchers suggest flows through the study area.

The acoustic characteristics of three fine-grained sediments (silty sand, silt, silty clay) in the South China Sea (SCS) were measured and analyzed at high frequency range of 27–247 kHz. The measurement results show that the sound speed dispersion is a positive linear relation at the measured frequency range, and the attenuation follows nonlinear frequency dependence, α = kfⁿ, where n ranges from 0.59 to 0.85 for the three different sediments in the SCS. The frequency dependence of sound speed and attenuation were compared with the published literature. It was found that for silty clay, clayey silt, silt, and silty sand, the dispersion characteristics of these four sediments are basically consistent; in general, the dispersion of coarse particles is significant, and that of fine particles is weak, and permeability is the key parameter that determines the inflection point of high frequency to low frequency. By modeling these sediments with the Biot–Stoll model, it was found that the Biot–Stoll model can better predict the frequency-dependent characteristics of sound attenuation in a high-frequency band under the matching constraints of sound speed dispersion characteristics, indicating that the Biot–Stoll model has good applicability to different types of sediments in a high-frequency band.